PeanutAD: A Real-World Dataset for Anomaly Detection in Agricultural Product Processing Line

Anomaly detection in production lines - automatically detecting defects in product processes - plays a vital role across several domains. Several industrial anomaly detection datasets have been built with severe challenges to overcome. However, the need for more amount and diversity of training data in specific systems in the real world, especially in agricultural product processing lines, remains to be considered. In this work, we introduce a peanut anomaly detection dataset (PeanutAD), a real-world dataset for anomaly detection tasks, reflecting the actual conditions in production lines. Experiments on this dataset demonstrated significant room for improvement because of some challenges. With PeanutAD, we aim to expand the anomaly detection database quantitatively and qualitatively, allowing researchers to further enhance and evaluate anomaly detection techniques with the most realistic setup possible. PeanutAD and experimental results on baseline methods shall be kept fully open-source as a contribution to the research community
¹
.